Billionaire Jeff Bezos came calling. These researchers walked away to build physics AI, leaving a $2 mn salary behind

Billionaire Jeff Bezos-backed Project Prometheus came calling with a multimillion-dollar offer and more than $2 billion in planned funding. Two AI researchers chose to walk away — and build an entirely different kind of AI instead.

Reuters reported, citing documents reviewed by its reporters, that AI researchers Anima Anandkumar and Benedikt Jenik were offered key roles in Prometheus, the AI venture co-founded by Bezos and biotech entrepreneur Vik Bajaj.

The proposal offered Anandkumar a role as the company's public face and scientific leader, while she and Jenik were offered a combined 35% stake and a joint annual salary of $1 million, rising to $2 million after three months.


The proposal also outlined more than $2 billion in committed funding through Series B, from investors including Bezos, according to the offer letter seen by Reuters.

But Anandkumar and Jenik chose to keep building their own company, Accelerated Understanding Inc, rather than join Prometheus.

On Tuesday, the duo unveiled what they say is a new AI model designed not to understand language, but to understand physics.

The system can process as many as 5 trillion pieces of data in a single prompt, according to the company. That is about 5 million times the amount of information that flagship models from Anthropic and Google can typically process, Reuters reported.

Instead of using the Transformer architecture that powers ChatGPT and many other leading AI systems, Accelerated Understanding uses a technology known as neural operators to predict physical phenomena across space and time.

“The language-centric view of intelligence is humans at the center. Putting physics at the center is a nature-centric view,” Anandkumar, a Caltech professor, told Reuters.

The company's pitch is that AI built around physics could tackle problems that traditional language-focused models struggle with.

Accelerated Understanding sees potential applications in chip design, robotics, extreme weather forecasting and energy, Anandkumar said. In chip manufacturing, for example, the company believes a deeper understanding of physics could help optimise materials and temperatures without relying on repeated laboratory testing.

The company is initially targeting enterprise customers rather than consumers, the co-founders told Reuters.

From Prometheus to a different AI bet

The decision comes as the AI industry races beyond chatbots and text-based models toward systems that can better understand the physical world.

Several companies, including startups linked to AI researchers Yann LeCun and Fei-Fei Li, are working on so-called world models. Accelerated Understanding is betting that neural operators can provide a more general way of modelling physical phenomena.

Anandkumar's path to the idea began at Nvidia, where she worked from 2018 and led scientists researching how graphics processing units could be used for advanced AI.

An early project demonstrated how AI could speed up weather prediction while matching the accuracy of complex traditional forecasting methods. Nvidia CEO Jensen Huang later highlighted her work on neural operators at the company's 2021 GTC conference.

“He just got so excited,” Anandkumar told Reuters, recalling Huang's reaction. She said Huang responded: “I want it to eat all their lunches.”

Anandkumar declined to disclose funding for Accelerated Understanding but said the company already has partnerships with computing providers that have supplied hardware clusters to develop and run its AI.

Meanwhile, Bezos and Bajaj continued building Prometheus, which raised $12 billion in a Series B in June 2026 and is targeting AI systems that can automate the manufacturing of complex physical systems.

For Anandkumar and Jenik, however, the bet is now different: rather than helping build Bezos' Prometheus, they are trying to build AI that can understand the forces of nature itself.